Spark版本定制:三. SparkStreaming 透彻理解三板斧之三-CSDN博客

感谢DT大数据梦工厂支持提供以下内容，
DT大数据梦工厂专注于Spark发行版定制。详细信息请查看
联系邮箱18610086859@126.com
电话：18610086859
QQ:1740415547
微信号：18610086859

IMF晚8点大数据实战YY直播频道号：68917580

以计算电商中各类别中最热门的商品排名的应用程序作为切入点，进一步

理解Spark Streaming Job整个架构和运行机制

代码如下：

package sparksql.java.sparkStreaming

import org.apache.spark.SparkConf

import org.apache.spark.sql.Row

import org.apache.spark.sql.hive.HiveContext

import org.apache.spark.sql.types.{IntegerType, StringType, StructField, StructType}

import org.apache.spark.streaming.{Seconds, StreamingContext}

/**

* 使用Spark Streaming+Spark SQL来在线动态计算电商中不同类别中最热门的商品排名，例如手机这个类别下面最热门的三种手机、电视这个类别

* 下最热门的三种电视，该实例在实际生产环境下具有非常重大的意义；

* @author DT大数据梦工厂

* 新浪微博：http://weibo.com/ilovepains/

* 实现技术：Spark Streaming+Spark SQL，之所以Spark Streaming能够使用ML、sql、graphx等功能是因为有foreachRDD和Transform

* 等接口，这些接口中其实是基于RDD进行操作，所以以RDD为基石，就可以直接使用Spark其它所有的功能，就像直接调用API一样简单。

* 假设说这里的数据的格式：user item category，例如Rocky Samsung Android

object OnlineTheTop3ItemForEachCategory2DB {

def main(args: Array[String]){

/**

* 第1步：创建Spark的配置对象SparkConf，设置Spark程序的运行时的配置信息，

* 例如说通过setMaster来设置程序要链接的Spark集群的Master的URL,如果设置

* 为local，则代表Spark程序在本地运行，特别适合于机器配置条件非常差（例如

* 只有1G的内存）的初学者 *

val conf = new SparkConf() //创建SparkConf对象

conf.setAppName("OnlineTheTop3ItemForEachCategory2DB") //设置应用程序的名称，在程序运行的监控界面可以看到名称

// conf.setMaster("spark://Master:7077") //此时，程序在Spark集群

conf.setMaster("local[6]")

//设置batchDuration时间间隔来控制Job生成的频率并且创建Spark Streaming执行的入口

val ssc = new StreamingContext(conf, Seconds(5))

ssc.checkpoint("/root/Documents/SparkApps/checkpoint")

val userClickLogsDStream = ssc.socketTextStream("Master", 9999)

val formattedUserClickLogsDStream = userClickLogsDStream.map(clickLog =>

(clickLog.split(" ")(2) + "_" + clickLog.split(" ")(1), 1))

// val categoryUserClickLogsDStream = formattedUserClickLogsDStream.reduceByKeyAndWindow((v1:Int, v2: Int) => v1 + v2,

// (v1:Int, v2: Int) => v1 - v2, Seconds(60), Seconds(20))

val categoryUserClickLogsDStream = formattedUserClickLogsDStream.reduceByKeyAndWindow(_+_,

_-_, Seconds(60), Seconds(20))

categoryUserClickLogsDStream.foreachRDD { rdd => {

if (rdd.isEmpty()) {

println("No data inputted!!!")

} else {

val categoryItemRow = rdd.map(reducedItem => {

val category = reducedItem._1.split("_")(0)

val item = reducedItem._1.split("_")(1)

val click_count = reducedItem._2

Row(category, item, click_count)

})

val structType = StructType(Array(

StructField("category", StringType, true),

StructField("item", StringType, true),

StructField("click_count", IntegerType, true)

))

val hiveContext = new HiveContext(rdd.context)

val categoryItemDF = hiveContext.createDataFrame(categoryItemRow, structType)

categoryItemDF.registerTempTable("categoryItemTable")

val reseltDataFram = hiveContext.sql("SELECT category,item,click_count FROM (SELECT category,item,click_count,row_number()" +

" OVER (PARTITION BY category ORDER BY click_count DESC) rank FROM categoryItemTable) subquery " +

" WHERE rank <= 3")

reseltDataFram.show()

val resultRowRDD = reseltDataFram.rdd

resultRowRDD.foreachPartition { partitionOfRecords => {

if (partitionOfRecords.isEmpty){

println("This RDD is not null but partition is null")

} else {

// ConnectionPool is a static, lazily initialized pool of connections

val connection = ConnectionPool.getConnection()

partitionOfRecords.foreach(record => {

val sql = "insert into categorytop3(category,item,client_count) values('" + record.getAs("category") + "','" +

record.getAs("item") + "'," + record.getAs("click_count") + ")"

val stmt = connection.createStatement();

stmt.executeUpdate(sql);

})

ConnectionPool.returnConnection(connection) // return to the pool for future reuse

}

ssc.start()

ssc.awaitTermination()

}

运行说明：

1、在StreamingContext调用start方法的内部其实是会启动JobScheduler的Start方法：

// 在新的线程里开始流式调度，在不影响当前线程的情况下，线程的局部属性如调用点和工作组将被重置。
ThreadUtils.runInNewThread("streaming-start") {
  sparkContext.setCallSite(startSite.get)
  sparkContext.clearJobGroup()
  sparkContext.setLocalProperty(SparkContext.SPARK_JOB_INTERRUPT_ON_CANCEL, "false")
  scheduler.start()
}

在JobScheduler的start内部会构造JobGenerator和ReceiverTacker，并且调用JobGenerator和ReceiverTacker的start方法：

receiverTracker = new ReceiverTracker(ssc)
inputInfoTracker = new InputInfoTracker(ssc)
executorAllocationManager = ExecutorAllocationManager.createIfEnabled(
  ssc.sparkContext,
  receiverTracker,
  ssc.conf,
  ssc.graph.batchDuration.milliseconds,
  clock)
executorAllocationManager.foreach(ssc.addStreamingListener)
receiverTracker.start()
jobGenerator.start()

2、ReceiverTracker启动后首先在Spark Cluster中启动Receiver（其实是在Executor中先启动ReceiverSupervisor），在Receiver收到数据后会通过ReceiverSupervisor存储到Executor并且把数据的Metadata信息发送给Driver中的ReceiverTracker，在ReceiverTracker内部会通过ReceivedBlockTracker来管理接受到的元数据信息

/** Start the endpoint and receiver execution thread. */
def start(): Unit = synchronized {
  if (isTrackerStarted) {
    throw new SparkException("ReceiverTracker already started")
  }

  if (!receiverInputStreams.isEmpty) {
    endpoint = ssc.env.rpcEnv.setupEndpoint(
      "ReceiverTracker", new ReceiverTrackerEndpoint(ssc.env.rpcEnv))
    if (!skipReceiverLaunch) launchReceivers()
    logInfo("ReceiverTracker started")
    trackerState = Started
  }
}

private val receivedBlockTracker = new ReceivedBlockTracker(
  ssc.sparkContext.conf,
  ssc.sparkContext.hadoopConfiguration,
  receiverInputStreamIds,
  ssc.scheduler.clock,
  ssc.isCheckpointPresent,
  Option(ssc.checkpointDir)
)

// Local messages
case StartAllReceivers(receivers) =>
  val scheduledLocations = schedulingPolicy.scheduleReceivers(receivers, getExecutors)
  for (receiver <- receivers) {
    val executors = scheduledLocations(receiver.streamId)
    updateReceiverScheduledExecutors(receiver.streamId, executors)
    receiverPreferredLocations(receiver.streamId) = receiver.preferredLocation
    startReceiver(receiver, executors)
  }

// Function to start the receiver on the worker node
val startReceiverFunc: Iterator[Receiver[_]] => Unit =
  (iterator: Iterator[Receiver[_]]) => {
    if (!iterator.hasNext) {
      throw new SparkException(
        "Could not start receiver as object not found.")
    }
    if (TaskContext.get().attemptNumber() == 0) {
      val receiver = iterator.next()
      assert(iterator.hasNext == false)
      val supervisor = new ReceiverSupervisorImpl(
        receiver, SparkEnv.get, serializableHadoopConf.value, checkpointDirOption)
      supervisor.start()
      supervisor.awaitTermination()
    } else {
      // It's restarted by TaskScheduler, but we want to reschedule it again. So exit it.
    }
  }

// Create the RDD using the scheduledLocations to run the receiver in a Spark job
val receiverRDD: RDD[Receiver[_]] =
  if (scheduledLocations.isEmpty) {
    ssc.sc.makeRDD(Seq(receiver), 1)
  } else {
    val preferredLocations = scheduledLocations.map(_.toString).distinct
    ssc.sc.makeRDD(Seq(receiver -> preferredLocations))
  }
receiverRDD.setName(s"Receiver $receiverId")
ssc.sparkContext.setJobDescription(s"Streaming job running receiver $receiverId")
ssc.sparkContext.setCallSite(Option(ssc.getStartSite()).getOrElse(Utils.getCallSite()))

3、每个BatchInterval会产生一个具体的Job，其实这里的Job不是Spark Core中所指的Job，它只是基于DStreamGraph而生成的RDD的DAG而已，从Java角度讲，相当于Runnable接口实例，此时要想运行Job需要提交给JobScheduler，在JobScheduler中通过线程池的方式找到一个单独的线程来提交Job到集群运行（其实是在线程中基于RDD的Action触发真正的作业的运行），为什么使用线程池呢？

1，作业不断生成，所以为了提升效率，我们需要线程池；这和在Executor中通过线程池执行Task有异曲同工之妙；

2，有可能设置了Job的FAIR公平调度的方式，这个时候也需要多线程的支持；

ReceiverSupervisor中：

/** Start receiver */
def startReceiver(): Unit = synchronized {
  try {
    if (onReceiverStart()) {
      logInfo(s"Starting receiver $streamId")
      receiverState = Started
      receiver.onStart()
      logInfo(s"Called receiver $streamId onStart")
    } else {
      // The driver refused us
      stop("Registered unsuccessfully because Driver refused to start receiver " + streamId, None)
    }
  } catch {
    case NonFatal(t) =>
      stop("Error starting receiver " + streamId, Some(t))
  }
}

在Receiver子类以

onStart方法

val executorPool =
  ThreadUtils.newDaemonFixedThreadPool(topics.values.sum, "KafkaMessageHandler")
try {
  // Start the messages handler for each partition
  topicMessageStreams.values.foreach { streams =>
    streams.foreach { stream => executorPool.submit(new MessageHandler(stream)) }
  }
} finally {
  executorPool.shutdown() // Just causes threads to terminate after work is done
}
  
// Handles Kafka messages
private class MessageHandler(stream: KafkaStream[K, V])
  extends Runnable {
  def run() {
    logInfo("Starting MessageHandler.")
    try {
      val streamIterator = stream.iterator()
      while (streamIterator.hasNext()) {
        val msgAndMetadata = streamIterator.next()
        store((msgAndMetadata.key, msgAndMetadata.message))
      }
    } catch {
      case e: Throwable => reportError("Error handling message; exiting", e)
    }
  }
}

转载于:https://my.oschina.net/DTSpark/blog/672491